Simon Willison

Large Language Models are the engines that power AI tools like ChatGPT, Bing and Bard. They are deeply weird, often frustrating and undeniably fascinating pieces of technology. This talk will attempt to summarize everything I've learned about them over the past year: how they are built, what they can do, what they can't do and how we can best tame them and use them to solve interesting problems. LLMs don't have to be large: I'll talk through the latest developments in openly licensed models you can run on your own devices, including on your phone and even in your web browser. I'll also discuss their as-yet unconquered security flaws, and the bewildering ethical implications of how they are trained and what they can damage.

Datasette is a tool for exploring and publishing data. It helps people take data of any shape or size, analyze and explore it, and publish it as an interactive website and accompanying API.

Datasette is aimed at data journalists, museum curators, archivists, local governments and anyone else who has data that they wish to share with the world. It is part of a wider ecosystem of tools and plugins dedicated to making working with structured data as productive as possible.

I’ll use this talk to introduce Datasette and show how it can be used to quickly explore and publish data.

I’ll talk about why SQLite is an ideal publishing format for structured data and demonstrate several open source tools for converting data from different sources into SQLite databases for use with Datasette.

Datasette is the foundation of my Dogsheep personal data warehouse project. I’ll show how I’ve pulled together data about my own life from a wide range of sources - from Twitter and GitHub to 23AndMe and Apple Health - and used it to build my own data warehouse to help me answer questions about myself.